Modelling Linguistic Causation

DSA ADS Course - 2021

Causal Data Science, Causal Inference, Linguistic Causation, Structural Equation Modelling

Discuss causative linguistic expressions, how causal relations are expressed in natural languages, and Structural Equation Modelling (SEM).

Modelling Linguistic Causation - 2021


This paper introduces a systematic way of analyzing the semantics of causative linguistic expressions, and of how causal relations are expressed in natural languages. The starting point for this broad agenda is to provide an explanation for the asymmetrical inferential relationship between two causative constructions: change-of-state (CoS) verbs and the verb cause, commonly ascribed to the former having an additional prerequisite of direct causation. The direct causation hypothesis, however, is fraught with empirical and theoretical challenges. At the theoretical level, capturing the felicity conditions specific to CoS verbs and the notion of direct causation requires a means of modelling complex causal structures. This is on no account a trivial task, as it necessitates, inter alia, modelling causation in a way that is germane to the linguistic expressions designating such relations. Hence, the main objective of this paper is to develop a framework for modelling the  semantics of causal statements. For this purpose, this paper makes use of the framework
of Structural Equation Modelling (SEM), and it demonstrates how this approach provides tools for a rigorous model-theoretic treatment of the differential semantics of causal  expressions. This paper introduces formal logical definitions of different types of conditions  using SEM networks, and show how this proposal and the formal tools it employs allow us to make sense of the asymmetric entailment relationship between the two constructions. In our proposal, CoS verbs do not require contiguity between cause and effect at all, but instead they require that its subject is set by default to a participant in completion event, the event which “completes” a sufficient set of conditions, such that following this event (but not before) the values of the set of conditions in the sufficient set entail that the effect occurs. According to this, the intuition of direct causation arises (epiphenomenally) from contrasting CoS verbs with overt cause sentences: the stronger selection pattern of the former - which requires a completion event - may exclude more temporally distant conditions, while the latter admits any necessary condition. 

Resource Type: